Transcripts
1. Direct AI Video Like a Cinematographer-Trailer: Hi. I'm Sagai Babu. I make software explainer
videos for living and I have tutored long enough to know
where people get stuck. Here's the problem. One A
clip looks like a film, put two side by side
and it falls apart. The sun jump side, the mug changes color, the plant grows new leaves. In this class, you
will fix that by treating every scene
as a frozen predict. Lock the word, move the lines. The voice you're hearing was enhanced for clarity
with Lamon laps. The short video clips you will see throughout we made
with Google Flow. Here's where you're stuck. You've made some
gorgeous AI clips. One on its own
looks like a film. Then you put two of them
next to each other, and the whole thing falls apart. The sun was on the left,
now it's on the right. The uk had a blue rim. Now it's plain white. The plant on the desk had seven leaves. Now it's got 12, and
it's a different plant. Your character's cardigan
turned from mustard to brown somewhere between
shot two and shot three. Nothing cuts together,
and you can't work out why because every single prompt looked fine when you wrote it. That's the exact problem
the next five modules fix. And the fix isn't
a better prompt. It's a different way of
thinking about the scene. We're going to treat every scene like a frozen three D set, a film set where nobody's
allowed to touch anything. The light is locked. Every prop has a count, a place, and a direction. The only thing that
moves is the camera. Lock the world, move the lens, and your clips start behaving
like shots in a real edit. I'm going to teach all
of this in Google Flow, which runs Google's
VO video models because that's where
I've been living lately. If you use runway or anything
else, don't close this. Everything here still works. The tool changes, the set
sheet doesn't the end, you'll have a six
shot sequence of your own built like a
cinematographer would build it. Right, let's get the desk ready.
2. Class Project: Class project. Your
class project is a six short sequence built
from one frozen set. You will make it in stages. Each module ends with a small project that becomes one piece
of the final built. In Module one, you first
go three rough images. In Module two, you set sheets. I Model three, six camera lines. I Model four, a
diagnosis of what broke. In Model five, you finished
sequence on a timeline. Share each piece in the
project gallery as you go. Raf is welcome wonky and
consistent is as sequence.
3. Module 1: Getting Set Up and Your First Go [Project included]: And Lesson 1.1. What's on the desk? Not much, honestly. Let me walk you around the desk. First, an account in
an AI video tool. I'm using Google Flow. Runway works fine, too, and so will most tools that let you start a clip from
an image you give it. That last bit matters. A tool that can only take text is harder to
keep consistent. So if yours lets you upload a starting image or
a reference image, find where that button lives
now before we need it. Second, an image generator. Think of it as your sketchbook
before the camera rolls. Some video tools have one
built in, and that's handy. If yours doesn't whichever
one you already know. We'll make key images first
and turn them into clips second because a still picture is cheaper to fix
than a moving one. It's a bit like a tailor pinning a shirt before
cutting the cloth. Third, a short
piece of voiceover. Two or three
sentences is plenty. If you haven't got
one, borrow mine, which we use all class. A new request lands at four. Asha opens the list and
spots it straightaway, a refund that's three days late. One click, one reply, and the customer has an answer
before the T goes cold. And last, a simple video editor for the final
module cap cut, DawnciRsolve, I movie, or the clip strip
inside your video tool. You only need to put clips
in a row and trim them. Lesson 1.2, one
folder, one window. The most important tool on
the desk costs nothing, a plain text document, Notsap, a Word file, Google Docs, whatever opens fastest
on your machine. This is where your set sheet and your short list will live. I keep mine open in a window on the right of my screen the
whole time I'm prompting, and I copy from it constantly
Prompt box on the left, text document on the right. You'll know the setups
working when you never type a scene description straight
into the prompt box again. Then make a folder and call it something boring
like Asha sequence. Every image and clip goes in
there named by shot number. Shot 01, shot 02, and so on. I once lost an
afternoon hunting for the one good clip among
40 called Download. Future you will
be very grateful. Project one, your first go. Before I teach you
a single thing, I want you to make something
badly. Here's the brief. Take the first two
lines of the voiceover, write three prompts for three
shots of the same scene. A woman at a desk by a window getting a new customer
request on her laptop. A white shot of the room, a medium shot of her, and a close shot of
the laptop screen. Generate the three key images or three clips if
your tools quick. Then put them side by side. I'm not going to tell you
how to write those prompts. Not yet. Write them however
you'd normally write them. I'm holding the method back on purpose because if you
find the problem yourself, the next hour stops being my answers and becomes the
answers to your questions. That's a completely
different class. You've got about 9 minutes. If generation is slow
on your account, write all three prompts first, then generate what you can. Even two images will
show you the problem, and it will come out rough. I promise you that. The
first time I tried this, I got three beautiful
pictures of what were clearly three different women
in three different offices. One had a window,
one had a bookshelf, one had a city skyline I
definitely never asked for. Mine was terrible.
Everyone's is. When you're done,
don't delete anything. Save the three images in
your folder as Project one. Then write one sentence at the
top of your text document. The one thing that bothered you most when you looked
at them side by side. Whatever it is, write
it down in plain words. That sentence is what this
class is about to fix. Three questions, and there are
no wrong answers to these. Have a think before I tell
you what most people say. First question, which changed more between your three images? The person or the room? Most people say both, and that's the honest answer. It's the first clue
that the tool was guessing at everything
you didn't write down. Second question. Look at
the light in each image. Can you say which side of
the room it came from? In most rooms I teach, nobody can say for all three. That tells you the light
was never decided. It was left to
chance three times. Third question, and this
one's a judgment call. If you could fix only one
thing across the three, which would make them
feel most like one scene? I pick the light
every time because the eye reads light
before it reads props. If you pick the room,
that's a fine answer too. Hang on to your
reason. Module recap. The first version
isn't the draft. It's the diagnosis. You've got three rough images and one sentence
naming what hurt, and that sentence is the thread through the
next three modules. Keep it where you
can see it because Module two starts
with why it happened.
4. Module 2: The Frozen Set [Project included]: Listen, 2.1. The
tool has no memory. Right, phones are
down. Look up here. Let's look at what you wrote
at the top of your document. I bet most of you
wrote something about light or something about
stuff on the desk changing. Those two come up
nearly every time, and they come from one cause. Here it is, and I want
you to really hear it. The tool has no memory
of your last clip. None. Every time you press generate, it starts from nothing. It reads your words and it fills every gap you left
with its own guess. You wrote a woman at a desk. It had to guess her hair, her clothes, the desk, the wall, the time of day, which
side the windows on, and how many things
are on the desk. Three prompts mean three
separate sets of guesses. No wonder they don't match. Nobody told them to. So the job isn't to write cleverer prompts. The job is to leave fewer gaps. Et me give you a
picture to hold. Think of a theater
stage seen from up in the balcony.
There's a back wall. There's a window cut
into the left wall, and sunlight falls through it in a long slanted rectangle
across the floor. There's a desk in
the middle with things sitting on
it in fixed places. Now, picture the camera as a little person walking
around that stage. It can stand close or far. It can crouch or climb a ladder. It can walk slowly
while filming, but it can't move the desk
and it can't move the sun. That's a frozen three D set. A real film crew works
exactly like this. They like the set once, dress it once, and shoot
it from many angles. Every angle matches because the world in front of the
camera never changed. Film people call that
matching continuity, and on a big shoot, there's a person whose whole
job is watching it. You're going to be that person. Lesson 2.2, the set sheet. The tool for this is the first
named move of the class. I call it the set sheet. Think of it as the
floor plan and props list a set dresser
carries on a clipboard. It's a short paragraph in your text document
that describes the frozen set so completely that the tool
has nothing left to guess. You write it once,
then you paste it into every single prompt word
for word without editing. Here's mine for Asha's
desk. I'll read it slowly. A small home office with
a plain pale green wall, one wooden desk
facing the camera, one window on the
left of the frame. Warm late afternoon sun at 4:00 coming in low
from the left. One silver laptop in
the center of the desk. One white mug with
a thin blue rim to the right of the laptop,
handle pointing right. One potted plant with seven leaves at the back
right corner of the desk. Asha, mid 30s with
short black hair, small round glasses, and a mustard cardigan
over a white t shirt. Photo realistic,
soft natural light, shallow depth of field, 35 millimeter film look. No logos anywhere. Notice what's in there. The
light has a source, a side, and a ti, not nice lighting, a window on the left at 4:00. That's three facts the
tool can't misread. Every prop has a number, a place, and where it
matters, a direction. If you don't say which way
the mug handle points, the tool picks, and it'll
pick differently every time. Asha has the same four or five
fixed details every time. And there's a style
line at the end, which we lean on in Module four. Let me do the arithmetic
because it's sobering. A desk scene like mine
has about ten things that can change
wall, window side, light angle, time of day, desk, laptop, mug, plant,
hair, and clothes. Leave all ten open
across six shots, and that's 60 separate guesses. You'd need every one of the 60 to land the same way by luck. Write them down and the
guesses drop to near zero. Three things trip
people up here. The first is a set sheet
that's all mood and no facts. Warm, inviting, cinematic tells the tool how
to feel, not what to build. Swap every mood word
you can for a fact. For example, Cozy becomes one knitted throw folded
over the chair back. The second is forgetting
the background. People lock the character
and leave the wall open. Then wonder why shot
three has a bookshelf. The third is the opposite. A set sheet so crowded the
tool starts dropping details. I keep background
props to four or five. Minimal sets are
easier to freeze. Lesson 2.3, the seven leaf rule. The second named move is the one I'm a bit
obsessive about. I call it the seven leaf rule. If the plant has seven
leaves in shot one, it has seven leaves in shot six. Counts don't change. Positions don't change,
directions don't change. Not unless you change
them on purpose by name. It sounds fussy. It isn't a continuity photo
on a film shoot. The snapshot someone takes, so the actor's collar sits
the same way after lunch. Viewers don't count leaves, but their eyes notice
something's off, even if they can't say what. That vague wrongness is what you felt looking at your
Project one images. The rule cuts both ways. If the storing needs the mug
to be empty in shot six, you write the same white mug
now empty in shot six only. You change it by name. Nothing ever changes
by accident. If you've been making
AI videos for a while, here's the bit I bet
you're getting wrong. You're probably rewriting
your scene description slightly in every prompt, because repeating
yourself feels lazy, a cozy office in one, a warm home workspace
in the next. To you, those mean the same. To the tool, they're
two different rooms. Stop paraphrasing. Paste. It's the same as a chef working from a recipe
card on a busy night. Nobody puts the salt on
the hundredth plate. Repetition is the whole trick, and it's the hardest
habit for good writers to accept. Try it. Write your set sheet. Right, pens down in 5 minutes. Write the set sheet
for your own scene or rewrite mine in your
own words for practice. Give the light a source, a side, and a time. Give every prop a
count and a place. Give your character
four fixed details. End with a style line. You're done when you read
it back and can't find a single word a stranger
could picture two ways. Quick quiz four questions. Shout them out in your head. Got them? Here's what
I'd say. First question. Why did your project
one image drifts? Because the tool has no
memory of the last clip. Every generation fills your
gaps with fresh guesses. Second question,
What three facts does light need in a set sheet? A source, a side, and a time. The window on the left at 4:00. Golden is a mood, not a fact. Third question, Your mug needs to be empty
in the last shot. How do you write that without breaking the seven leaf rule? You change it by name
in that shot only. The same white mug now empty. The rule bans accidents,
not decisions. Fourth question, a judgment one. Your set sheet has
11 background props, and the tool keeps
dropping two of them. Add more detail or cut props. Cut. A crowded
sheet gets skimmed, and four or five props
you can hold beat 11. You can't. Module recap. The set never moves.
The camera does. Your set sheet
freezes the world, and the seven leaf
rule keeps it frozen. The frustration
you wrote down in Module one was mostly
a missing line. Now that the set's locked, Module three hands you the one thing you're
allowed to change.
5. Module 3: The Camera Pass [Project included]: Listen, 3.1, size,
angle, and one move. We'll spend about
20 minutes here. This is the heart of the class. When I first made explainer
sequences this way, I kept everything locked and
still got board footage. Six shots, all the same, polite medium shot, all
from the same height. It matched beautifully. It was also completely dead. A student of mine
said it looked like a security camera that
had learned to act. She wasn't wrong. Locking
the set solves the matching. It doesn't solve
the storytelling. For that, you need the camera. That's the third named move. I call it the camera pass. Between one shot and the next, you keep the set sheet identical and change
only the camera. Three things about
the camera can change how big the
subject is in the frame, the angle you see it from, and how the camera
moves during the clip. Every prompt from now on is the set sheet plus
one camera decision. Short size is how much of
the world fits in the frame. A wide shot shows the whole room with
Asha small inside it, the window and the slanted
light all visible. It tells the viewer
where we are. A medium shot shows her
from about the waist up, and it's where most
conversation lives. A close up fills the frame
with her face or one object. An extreme close up
goes tighter still, her eyes or just the
blue rim of the mug. Think of a football broadcast. The white shot shows the whole pitch so you
understand the game. The close up shows the striker's
face before the penalty. You need both in
the right order. The angle is where
the camera stands. Eye level is neutral and honest. A high angle looks
down and makes the subject feel
smaller or watched. A low angle looks up and
makes it feel bigger. And one you'll use
constantly in explainer work is the over the
shoulder shot or OTS. The camera sits just
behind Asha's shoulder. We see the edge of her
hair and cardigan soft and blurred in the foreground and her laptop screen
sharp beyond it. It puts the viewer in her chair. Movement is what the camera
does during the clip. Think of the camera as
your own head and feet. A static shot doesn't move
at all, and it's underrated. A push in moves slowly towards the subject and
says, pay attention. A pull out moves away and
reveals more of the world. A pan turns left or right from a fixed spot like
turning your head. A tilt turns up or down like looking from the
desk up to the window. A dolly or tracking shot moves the whole camera
sideways as if on rails, kind of like filming out
of a slow train window. An orbit circles the subject. Handheld adds a
slight human wobble. My rule of thumb, and I'll
fight anyone on this. One camera move per clip. My clips usually come
out around 8 seconds, and there's room for
one clear move in that. Ask for a push in and a pan and an orbit, and
you'll get mush. A sentence with three verbs
is a sentence nobody follows. Vague words fail too. Dynamic angle means
nothing to the tool. Low angle, slow push in
means something. Lesson 3.2. If the voice says it,
the camera shows it. How do you choose
which camera decision goes with which shot? Not from your imagination,
from the voiceover. It's a bit like a radio
presenter describing a match where every sentence
tells you where to look. Read the line. Find the thing the listener needs to
see at that moment. Then pick the shot
that shows it best. Take a new request
lands at four. The listener needs to see
the request arriving. So the hero isn't Asha's face. It's the laptop screen. I'd go over the shoulder static, with a new row appearing at
the top of a ticket list. Take Asha opens the list
and spots it straight away. Now they need the list,
then the one ticket. Take before the T goes cold, they need the Notice two of those three shots
don't need Asha at all. Beginners put the character in every frame because they've
worked hard on her. Don't the voice talks
about the product, the product screen is the hero. Save her face for the
moments that need a person. And when the screen is the
hero, describe it yourself. Never let the tool
invent a screen. It'll happily write
gibberish words, scatter broken icons, and slap a made up logo
in the corner. So I describe it like a stencil. A plain ticket list on a
white background. Five rows. Each row has a customer name, a short subject line, and a small colored status tag. The top row is new
and highlighted in pale yellow with a
simple envelope icon. No logos. Where there's text, I write the actual words. One more habit from my day job. When you want to show
one specific item, show the whole list first and then move to that
tickets detail view. Jump straight into the detail, and viewers have no
idea where they are. It's the same reason a map app shows you the city before
it zooms to the street. That gives a six shot
sequence its shape. Shot one is the title shot. Think of it like the
cover of a book. Calm, clean, often the
set with nobody in it. Space for the title.
Shots two and three are mine, not the scripts. A wide establishing
shot of the room with a low 4:00 sun across the floor. Then a medium on Asha settling into her chair
on a slow push in. Only from shot four does
the script take over. Two shots of room
and person cost about 12 seconds and buy
you the whole audience. Lesson 3.3, the cut in pair, the last named move, and it's the one that
makes your edit breathe. I call it the cut in pair. Every shot gets two
images, a key image, which is the main shot
and one cut in or B roll. A short related shot you
can cut to and back. A cut in usually goes tighter on a detail inside the same set. In this class, role always comes from
the same frozen set. Each one gets its
own camera decision. Say shot six is a
medium close up of Asha typing her reply,
static eye level. It's cut in is an extreme
close up of the white mug with the blue rim and steam rising on a slow push in. Why the mug? Because the voice says before the t goes cold, different size, different movement,
same set sheet, same light from the left, same handle pointing right. Why bother? Cutting
between two angles of one moment is what makes footage feel edited instead
of generated, and a cut in is your safety net. If the main clip
has one bad second, you cut to the mug and
nobody ever knows. Here's the sharp aversion for anyone who's
done this before. Every move should be motivated, which means it has a
reason inside the story. A push in when the important
ticket appears is motivated. A push in because it looks
cinematic is decoration, and viewers feel the difference even when they can't name it. Before you add a move, finish this sentence out loud. The camera moves because if you can't finish
it, go static. Two more mistakes quickly. People change the set
while changing the camera, writing a new angle and quietly dropping the
mug from the prompt. Then the mug vanishes. Change the camera line only. And people forget which way the light falls once
the camera turns. If the camera looks back
towards the window, the light now comes towards
us. That's still correct. The sun didn't move,
the camera did. Say so. The window is now behind Asha back
lighting her hair. Try it. Six camera lines. Right, pens down in 6 minutes. Take your voice over or mine and write the camera
line for six shots. Shot one is titled shot one. Shots two and three are
for place and character, and shots four to six are
from the voiceover lines. Each gets a size, an angle and one move. Then add a cut in to each with
a different size and move. Words only, no generating. You're done when every
shot has two camera lines, and you can point the voiceover
words behind each one. Four questions. Have a
go before I tell you. First question. What are the three things about the camera you're allowed to change? Size, angle, and movement. Everything else lives
in the set sheet and stays put. Second question. The voice says the report
lands in her inbox. Who's the hero of that shot? The screen, not the person. The voice is talking
about the inbox, so the camera shows the inbox, probably over her shoulder. Third question, why show the ticket list before
the single ticket? Because viewers need to know where they are
before you zoom in. List first, then detail, like the city before the street. Fourth question, a judgment one. You've written slow push in while panning left
for shot five. Keep it or change it. Change it. That's two moves in one clip, and the tool will
blur them both. Pick the one you can finish. The camera moves because
four. Module recap. If the voice says it,
the camera shows it. The camera pass changes only
size, angle, and movement, and the cut in pair
gives every shot a partner from the
same frozen set. You've now got a full
shot list on paper. Module four is what happens when you press generate
and it fights back.
6. Module 4: Fixing What Breaks [Project included]: Lesson 4.1, light,
props or camera. Even with a perfect set
sheet, things break. They broke for me this week. I was making a sequence for
a software walk through, six shots. Every prop counted. Shot five came back
with the plant on the left side of the
desk. Same plant. Seven leaves I
counted. Wrong side. I'd spent a good
while regenerating before I realized the
plant line was fine. My camera line said
reverse angle, and the tool had flipped
the whole set to match. When a clip won't cut, there are only
three things it can be light, props or camera. Name the right one, and
the fix is usually small. Guess wrong, and you
burn an afternoon. First, a habit. Never judge a clip on its own. A clip alone almost
always looks fine, which is how you
got into this mess. Put it on the timeline next to the shot before it
and play the cut. Watch the last second of one and the first
second of the next. Your eye catches
in half a second what it would never
catch in a still. I call this the
back to back test, and I run it every time
a new clip arrives. When something jumps,
pause right on the cut and ask the three
questions in order. Say you've got a clip where the sun suddenly
sits on the right. Light jumps are the most common, and they nearly always come from light written as a mood
instead of a fact. Warm golden light is a feeling. The tool can satisfy
it from the left, the right, above, or
from a lamp it invented. The fix is the
clock and the side pasted identically
into every prompt. The other cause is
time of day drifting. You said afternoon in one
prompt and nothing in the next, and it came back at noon
with flat overhead light. If props drift, check
your set sheet first. Let's say the plant grew
to 12 leaves in shot five. Nine times out of ten, the prop line got
edited, shortened, or dropped, pasted
back in whole. If the sheets intact
and props still drift, give the tool a picture
instead of words. In flow, I make the
key image first. Check every prop against
the set sheet and only then turn that
approved image into a clip. Flow also lets you keep
reference images of your character and objects
and reuse them across shots, and runway has
reference images, too. A reference image is the seven
leaf rule in picture form, the way you'd hand a barber a photo instead of
describing a haircut. Character drift is props
drifting wearing a cardigan. For instance, her glasses go from round to square
between two shots. Same fixes, fixed details
in words every time. Then one reference image
of her re used for every shot and keep the
costume boring and specific. A busy pattern shirt is a gift
to the tool's imagination, and you don't want it imagining. Lesson 4.2, borrowed B roll
and the 180 degree line. Here's one people don't expect. The main shots match, but the cut ins look like stock footage from
a different film, sharper, colder, a
different lens feel. That's because cut
in prompts get written in a hurry and
lose the style line. The fix, cut ins carry the full set sheet and the same style line as
the character shots. Photorealistic,
soft natural light, shallow depth of field, 35 millimeter film look. A Bro shot is just the
camera walking closer. And then there's
my flipped plant. Camera problems usually
come from asking for too much or from
crossing the line. This is the bit for those of
you who've done this before. Think of a cricket
broadcast where the main cameras sit on
one side of the ground, so the batter always runs the same way
across your screen. Picture a straight line on the floor between
Asha and her laptop, running from her
chest to the screen. Film crews call it
the 180 degree line. As long as every camera
stays on one side of it, Asha always faces the
same way on screen. Say she's looking towards the right of the
frame in shot three, cross the line in shot four, and suddenly she's looking left and the viewer's
brain hiccups. That's what reverse
angle did to me. Pick a side of the room for
your camera and stay there. If you do cross on purpose, put a neutral shot in between, like a straight on screen
shot to reset the eye. That one's worth
sketching in your margin. A rectangle for the room, a dot for Asha, a
dot for the laptop, and a line joining them. Then a big half
circle on one side, which is where your
camera's allowed to stand. 10 seconds, no ruler. Last thing, when a clip fails, people regenerate
the same prompt over and over, hoping don't. Each try costs credits and time, and the same gaps produce
the same guesses. Name the problem,
fix the one line that caused it, then
regenerate once. If a clip is 90% right, keep it and cut away to its
cut in for the bad second. That's what editors
have always done. It's the same as a
magician's misdirection. You look at the mug and the
wobble slides past. Try it. Diagnose your first go. Pens down in 4 minutes. Go back to your project. One image and play detective. For each jump you see, name it. Light, props or camera. Then write the one line
you change to fix it. You're done when every jump has a name and a fix next to it. Four questions. Got your
answers. Here's mine. First question. What are the three things a
broken cut can be? Light props or camera. Ask them in that order, paused right on the cut. Second question, Your cut ins look colder and sharper
than your main shots. What went missing?
The style line. Same set, same light, same style, or the B
roll looks borrowed. Third question, Asha faces right in shot three
and left in shot four, and nobody moved
her. What happened? The camera crossed
the 180 degree line. Bring it back to its side of the room or put a
neutral shot between. Fourth question,
the judgment one. A clip is perfect, except for 1 second where
the mug handle flips. Regenerate or keep. Keep it, and cut to the mugs
cut in for that second, regenerating costs credits and might break something that's currently right. Module recap. A clip is a take, not a film. You're the one who decides
which takes make the cut. Run the back to back test, name the problem as light, props or camera
and fix one line. You now have everything
the main project needs. So Module five is
you working alone.
7. Your Main Project: Lesson 5.1, the brief and
what each stage uses. Everything up to now was
me talking you through it. Now you're the director. Now you're no, you're
talking. Here's the brief. Build a six shot sequence
for a short voiceover, about 20 to 30 seconds long. For every shot, write
a key image prompt, a cut in prompt,
and a video prompt. Generate the key images, check them and turn as many as your credits
allow into clips. Then put them on a
timeline in order. You start with the set
sheet from Module two. Count your props out loud. Give the light a source,
a side, and a time. Then the shot list,
which is Module three. Title shot, two shots
for place and person, three shots from the voiceover, each with one size, one angle, and one move, each
with its cut in pair. Then you generate,
which is Module four. Key images first checked against the set sheet before you
spend anything on video. Count the leaves,
check the handle, check which side the
light falls from. Then clips with the back to
back test as each one lands. Every prompt is
built the same way. Paste the set sheet,
add the camera line, add what happens taken
from the voiceover. So my shot four reads like
this, the full set sheet. Then over the shoulder
shot from behind Asha's right shoulder
static eye level. Her laptop shows a plane
ticket list with five rows. A new row slides in at the top, highlighted in pale yellow with an envelope icon and the subject line
refund not received. Lesson 5.2, making it yours
and where people stall. You can use Asha's desk
and my voice over, and that's a perfectly
good choice. Or pick your own scene, a baker opening a shop at dawn, a gardener on a balcony, a student at a library table. The one rule is one location and one character, so
you can freeze it. Two people in this
class should hand in completely
different sequences. If you've done this
before, raise the bar. Try one shot that crosses to the window side on purpose with a neutral shot in between. People stall in three places. The first is the set sheet
where they keep adding props. Stop at four or five. The second is shots
two and three, because they're
not in the script. Ask what a stranger walking
in would notice first, the light, the room, then the person.
That's your two shots. The third is the one clip
that refuses to behave. Diagnose it, fix one
line and try once more. If it still won't, keep the best version and plan
to cut it to its cut in. Here's what you're aiming at. A good one plays all six shots and nothing pulls your eye. A rough one has the right
shots in the right order. With a jump or
two, you can name. Both count is finished.
The main build. You've got about 12 minutes. So here's how I'd
spend them by feel. Three on the set sheet, three on the shot list,
and it's cut ins. The last six or so
generated key images, checked them, and made clips
from the approved ones. You're done when there's a
short list with six shots, three prompts each, and a timeline that
plays them in order. If credits run out, key images in order on a
timeline still count. Plenty of professional
storyboards are exactly that. Go. Quick quiz. The questions. And these
are about you, not me. Play your sequence
from the start first. First question, how many
cuts pulled your eye? If it's most of them,
you're still learning, and that's fine, because you
can name every jump now. A week ago, you couldn't. If it's one or two,
you're getting there. If it's none, you've got it. Second question, pick any shot. Can you point at the words in the voiceover that chose it? Still learning means
some shots were chosen because they
looked nice. Got it? It means every shot has a
reason you can say out loud. The third question
is the judgment one. If you had credits for
one more generation, which shot would you redo and
what line would you change? If you can answer
with a specific line, you've learned the most
important thing in this class. You fix sentences, not luck. Module recap. Now, put your sequence next to
your project one images. Same scene, St. Two, St. About an hour apart. One looks like three
strangers in three offices. The other looks like a film. That's what the hour
did. And if yours is still a bit wonky
in places, good. So was my first proper one. Wonky and consistent
is a sequence. Beautiful and random
is a pile of clips.
8. Closing: Let's step back. At the start, I promised you'd make
six clips look like they came off one set
and you've just done it. You wrote a set sheet that leaves the tool
nothing to guess. You held the seven leaf rule. You ran the camera pass, changing only size,
angle, and movement. You paired every
shot with a cutting, and when a clip
broke, you named it. If you remember only one
thing, remember this. The set never moves,
the camera does. Three things to make next, each a little harder. First, add six more
shots to the same set, so you're cutting a full
minute from one frozen world. Second, build a brand new
set and character from scratch and notice how much faster your second set
sheet comes together. Third, try two locations, say an office and a cafe
with two set sheets and one character who has
to stay identical in both. That's where this gets
properly interesting. To keep improving on your own, watch films with
the sound of for 10 minutes a week.
Count the shots. Notice where the camera
stands and when it moves. 10 minutes a week is
about 8 hours a year of studying real cinematographers
for free, on your sofa. You'll start seeing
shot lists everywhere. Thanks for staying with me.
Go and direct something. How I built this course. I built this course
around one frustration, AI clips that look gorgeous alone and fall
apart side by side. I assumed you'd already made a few clips in a tool
like flow or runway, and you're comfortable
typing a prompt. I also assumed nobody had ever taught you how a film set
or a short list works. I anchored everything in Google Flow and one
fictional six shot scene at Asha's desk because a
single concrete set makes every rule checkable. Tool features change quickly, so I kept the teaching
on things that hold in any tool that can
start from an image. For a 50 minute class,
keep modules two, three, and five and shorten
modules one and four.
Sahaya Babu Manickam, Motion Designer & Watercolor Artist
